You are on page 1of 26

Lecture 4

Noise as a random number generator

Plan of the Lecture


Extracting randomness from physical noise 1 Technical definition of noise 2-3 Thermal noise 4 Randomness from noise

MATHEMATICS OF NOISE

Many frequencies

Frequency spectra

Adding frequencies
1 1

0.8 0.8

M=40 M=10

0.6 0.6

0.4 0.4
0.2 0.2

0
-0.2 -0.2

-0.4 -0.4

0
0 1000 2000 3000 4000
5000 5000

-0.6 -0.6

-0.8 -0.8

6000

7000 7000

8000

9000 9000

10000 10000

Noise
More frequencies in the spectrum more noisy signal
1 0.8

Little correlation (short memory) hard to predict

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

In fact, many natural source of noise have a continuous spectrum:

Spectral bandwidth vs. memory time


Correlation time Coherence time Spectral bandwidth in frequency

THERMAL NOISE

Theory

A resistor

Really zero?

Contact with a thermal bath

Temperature T

Thermal noise

V
time 0

V=RI=0 is the average

Thermal fluctuations

Johnson-Nyquist noise
The thermal noise of the resistor is named after John Johnson, who reported it, and Harry Nyquist, who did the theoretical description (Bell Labs, 1932) From Johnsons paper: The mean-square potential fIuctuation over the conductor is proportional to the electrical resistance and the absolute temperature of the conductor. It is independent of the size, shape or material of the conductor.

The formula
Cable = 1D waveguide for E,B Bandwidth of the receiver
Power of thermal fluctuations

Power dissipated by the resistor

Boltzmanns constant

At equilibrium, they must be equal

More rigorous derivation

Predictions in graphs
Frequency spectrum

T=300K, Df=10kHz
1.6 10-13V2 =(0.4 mV)2

1kW

NOISE IN THE LAB

Devices
Oscilloscope

Amplifier

time

Observation

RANDOMNESS FROM NOISE

Statistics of noise

The observed distribution is enough, because we have a controlled source of noise: we checked that the noise is the one expected for thermal noise we trust physics that it is a phenomenon too complex to predict. One could put up an antenna and capture unknown signals, that may look random to us. But they may not be random for others, or may have unwanted structure in them.

Observed distribution
R=1kW T=300K

time

Counts (103)

Possible processing (1)


Try to get directly fair coin sequences out of it

00

0 1 1 0

11

More than one bit need to adapt carefully the intervals in order for each sequence to have the same probability

Possible processing (2)

Remark: for a trusted source, one can be less conservative and compute randomness from the average (Shannon) entropy.

Correlations from finite bandwidth


Previously we assumed that each value of V is independent of the others, but:

As we know, no problem in principle: just estimate P of each sequence, then compute Hmin. But in practice, this is not feasible.

Suggested Readings
Wikipedia pages: http://en.wikipedia.org/wiki/Hardware_random_number_generator

Summary of Lecture 4
Extracting randomness from physical noise

Noise = large spectrum of frequencies Thermal noise Extracting randomness, effects of correlations

You might also like